Curvilinear Effects of User-Generated Content on Hotels’ Market Share

Abstract

The present study is designed to expand the current literature in the effects of User-Generated Content (UGC) on firm performance, specifically to delineate the unique effects of the UGC components on market share results using a dynamic generalized method of moments (GMM) model. A longitudinal panel-data sample of 138 hotel reviews and ratings, along with monthly market share and several control variables is used for the empirical modeling. Overall, the analysis reveals that UGC has a positive impact on market share. A curvilinear mechanism explains the relationship between ratings and market share, discouraging lower-tiered firms from seeking a high rating. Moderating factors were also found to diminish the impact of a review length on market share. Theoretical explanations and managerial suggestions are offered in the discussion section.

Keywords

User-Generated Content ratings and reviews WOM trust market share dynamic model

Introduction

We live in a service economy, and the growth of the Internet has accelerated the trend away from brick-and-mortar traditional business models toward click-and-mortar models (e.g., buy at BestBuy.com and pick up at the store) or toward pure Internet-based models (e.g., Amazon.com). For the service industry, relying on the “mortar” part of the equation to deliver the service purchased, the Internet represents a great medium for communication and distribution. Yet, at the same time as the Internet gives more visibility to the service firms, it also gives more visibility and power to the user-consumer by empowering him or her to share postpurchase opinions and experiences with millions of potential viewers. This User-Generated Content (UGC) takes the form of any material created and uploaded to the Internet by nonmedia professionals, whether it is a comment left on Amazon.com, a video uploaded to YouTube, or a profile on Facebook (Interactive Advertising Bureau 2008). UGC constitutes a form of online customer engagement resulting in the shaping of the relationship between brands and consumers. UGC is a form of many-to-many communication, affecting not only the brand-customer relation but also the competitive landscape and consumers at large (Hoffman and Novak 1996). In some instances, UGC can be devastating for the firm (Kirkpatrick and Roth 2005) because of its multiplicative negative word-of-mouth (WOM) effect (Bansal and Voyer 2000).

Potential service clients concerned about their prepurchase decision increasingly turn to what are called “third party” websites (e.g., Tripadvisor.com, Zagat.com, Yelp.com, Angieslist.com) for reassurance. User reviews and ratings on services such as hotels, airlines, car rentals, restaurants, plumbers, doctors, or movies help consumers make decisions. These reviews are perceived as unbiased since they are posted by the service provider’s past customers, who appear to trust websites, such as Tripadvisor.com, to make their decisions (Verna 2008). These reviews can be used as a proxy for WOM (Dellarocas, Zhang, and Awad 2007), and modeled to improve revenue forecasts (Dellarocas, Zhang, and Awad 2007; Liu 2006). Some researchers, however, have found no effects between ratings and sales (Chen and Xie 2008; Duan, Gu, and Whinston 2008a), although the observed effects could be confounded by a between segment or product category effect (Zhu and Zhang 2010).

Despite these contrasted results, some firms have been shown to strategically manipulate online reviews in an effort to influence consumers’ purchase decisions (Dellarocas, Awad, and Zhang 2006). A recent user reviews and ratings report from TripAdvisor.com titled “2010 Dirtiest Hotels Lists” created an uproar in Europe. Published in the popular press, the report caused a group of hotelier to ask the government for regulation of third-party sites and UGC. A British newspaper reported that hotels throughout Europe were “seeking to persuade the European Union Commission to overhaul the rules governing web site reviews to ensure that they have been posted by genuine guests and not by rivals or people simply out to cause mischief” (Sharkey 2010). Hotel managers now monitor online critiques to improve service and “listen-in” on the online conversation using sophisticated algorithms in an attempt to preserve their market share (Yu 2010). Meanwhile, the same paper claims that consumers seem to be able to accept, or understand that a 70% rating is “good enough.” In fact, the latest American Consumer Satisfaction Index rates the average hotel industry satisfaction score at 77% (ACSI 2011), while J.D. Power’s hotel guest satisfaction averages 76.4%.¹

Current findings in the tourism and hospitality literature show support for a relationship between the postconsumption behaviors: satisfaction, WOM, and loyalty (Bronner and de Hoog 2010; Simpson and Siguaw 2008; Susskind, Bonn, and Dev 2003; Xie et al. 2011), while other research links WOM to purchase consideration and sales (Vermeulen and Seegers 2009; Ye, Law, and Gu 2009). While some findings show online reviews as less credible to consumers (Cox et al. 2009), others see a link to trust (Dickinger 2011) and behavioral intent (Stringham, Gerdes, and Vanleeuwen 2010).

In their recent review of electronic WOM (eWOM), Litvin, Goldsmith, and Pan (2008) call for future research related to the application of eWOM strategies; particularly to studies that will measure cognitive, affective, and behavioral implications upon consumer behavior and the new dynamics created by eWOM. Others have suggested the need for the study of the effects of UGC within and between different service segments of the same category (Zhu and Zhang 2010). Furthermore, Anderson and Mittal (2000) mentioned the importance of understanding the nonlinearity inherent in the service evaluation-consumer outcomes chain. Other researchers also point to the lack of research in this area (Vlachos, Pramatari, and Vrechopoulos 2011).

Thus, the aim of the present paper is to extend the literature on the nonlinear and dynamic relationship between UGC and the hotel performance, and more specifically, to demonstrate the detailed impact of each component of UGC on hotel market share. Through this research, we aim to make several contributions. First, our study advances the literature on UGC by including the dynamic effect of competition. Second, using a unique opportunity of a cross-segment panel data set we extend the current literature by modeling the effects of UGC within and between different service segments of the same category, thus answering researchers’ call (Zhu and Zhang 2010). Finally, the managerial implications of the present study might shed light in support of the recent trend for brands, such as Marriott or Hilton, to volunteer consumers’ reviews and ratings directly on their website despite the fear associated with this transparency (De Lollis 2012).

Conceptual Model and Hypotheses

A burgeoning area of the UGC spectrum is customer reviews and ratings providing an important source of information for the consumers (Chevalier and Mayzlin 2006; Jeong and Jeon 2008). Third-party firms such as Tripadvisor.com, Yelp.com or Angieslist.com, who offer a platform for UGC as their core offering, play an essential role in the competitive dynamic. Consumers can now share their opinions, complaints, and recommendations concerning products and services online, thereby engaging in eWOM behavior. The result of the rise of these third-party websites, along with the distribution websites (i.e., Orbitz, Expedia, Amazon, OpenTable) offering UGC and inventory in one place, is that consumers have an increasing amount of information available to them in order to make purchase decisions (O’Connor 2010; Sparks and Browning 2011). Chen and Xie (2008) suggest that these online customer communications may have a significant influence on product search and product choice. Other studies have found that consumers are more likely to trust brands that offer user ratings and reviews (Chevalier and Mayzlin 2006). In that respect, UGC, in the form of reviews and ratings, can be associated to eWOM behavior (Chen and Xie 2008), and used as a proxy (Dellarocas, Zhang, and Awad 2007). eWOM communication often occurs as part of the external search activities undertaken by consumers in their decision-making process (Sénécal and Nantel 2004). However, trust is necessary for consumer reviews to be effective as decision-making aids (Smith, Menon, and Sivakumar 2005). The eWOM receiver faces greater difficulty in judging the credibility of online product reviews, since the antecedents that are common in traditional WOM to determine information credibility, such as source similarity, expertise and accessibility, are not very appropriate in the online context (Feick and Higie 1992). Any evaluation of reviewer credibility or helpfulness is likely to come from the reviews themselves. Furthermore, differences between consumers’ motivation for engaging in eWOM have been shown (Bronner and de Hoog 2010). Individuals with different characteristics view online information seeking and online transactions differently. For example, consumers’ apprehensiveness about the Internet leads them to seek less information online and to limit their purchasing behavior (Susskind, Bonn, and Dev 2011). On the other hand, a well-known phenomenon in the tourism and hospitality industry describes a “billboard” effect, and suggests that online consumers might not all book through online travel agencies (OTAs), but are still seeking information on the web site in order to reach a decision (Hood 2011).

Most of the research on the relationship between WOM and trust places trust as a consequence of WOM (De Matos and Rossi 2008). Other researchers have explored the mediating role of trust in the link between eWOM and brand commitment (Bansal, Taylor, and St. James 2005; Ha 2004). Sparks and Browning (2011), for their part, have shown that a relationship exists between trust, the valence of the reviews and the purchase behavior of the consumer. Furthermore, consumers have been known to demonstrate a level of trust in using products’ reviews (Riegner 2007). However, reviews and ratings of services might suffer from the subjective nature of the assessment and judgment made by the reader (Laczniak, DeCarlo, and Ramaswami 2001; Folkes 1988; Sen and Lerman 2007; Weiner 2000). Research also shows that a weak relationship exists between the third-party consumer and the consumer posting the ratings and reviews (Bansal and Voyer 2000). It also shows that the perception of risk in a website transaction is perceived to be greater than a traditional face-to-face service encounter given the level of technology readiness (Zeithaml, Parasuraman, and Malhotra 2002), the amount of negative information posted (Ha 2004), and the complexity of the transaction (Harris and Goode 2004).

Conceptual Model

In this study, we conceptualize UGC as a bundle of two distinct but related components: ratings and reviews. Ratings are the numeric evaluations of the service satisfaction using a scale (often stars or points), and reviews are attached to the evaluations by the customers in the form of a narrative of different lengths and different tones (positive or negative) expressing a sentiment. Figure 1 depicts our conceptual model. It integrates temporal relationships between the UGC components, reviews and ratings, and market share. It also shows the within time period dynamic nature between market share and UGC, controlling for other factors such as location and segment, as well as the between time period correlations.

Figure 1.

Conceptual framework: user-generated content and hotels market share.

The uniqueness of the present research lies in the fact that actual hotel market share will be modeled as being dependent on UGC, thus extending Ye, Law, and Gu’s (2009) findings that demonstrate the positive impact of online reviews on hotel sales using the number of reviews as a proxy for hotel bookings.

The Dynamic Link between UGC and Market Share

There is strong evidence for the effect of UGC on companies’ sales. For instance, in a comparison of Amazon.com and BN.com, two online book sellers, Chevalier and Mayzlin (2006) found that the addition of a new favorable review on one site resulted in an increase in the book sales at that website relative to the other one. Others found that online user-generated movie eWOM volume at Yahoo.com significantly affected aggregate and weekly box office revenues (Liu 2006). Similar results have been observed in the hospitality industry (Ye, Law, and Gu 2009).

These studies have come short of integrating the effect of competition. Competition in the hotel industry is bounded by the location in which the service will be delivered. Hotel capacity is limited to its number of rooms, and its demand is location driven. Thus, local conditions will influence consumers’ choice, and a cheaper hotel located in another town has a low probability to be part of a consumer’s consideration set. Therefore, the effect of UGC on hotel sales is dependent on the competitive setting. For instance, a market in which only one hotel is present might behave differently than a market in which 20 hotels are competing. In the former case, the management might be able to lower the quality and increase the price to take advantage of the setting.

Along with the extensive WOM research body, we believe that alternative attractiveness, or the attraction to the alternative service providers (Bansal et al. 2005), needs to be accounted for. Previous self-reporting studies have found that alternative attractiveness is an important variable affecting customer decision (McDougall and Levesque 2000; Jones, Mothersbaugh, and Beatty 2000). Thus, an econometric model relying on time series data has to include the effect of the alternative choices (i.e., the competition) by modeling the dynamic relationship between a particular observed hotel performance and that of its competition.

Once the competition is added to the model, the relationship between ratings of past customers and the trust current customers experience reading these ratings can be dynamic (see Figure 1). In other words, a customer might patronize the market leader because she or he trusts the ratings provided on the website (Bansal, Taylor, and St. James 2005; Ha 2004; Riegner 2007), making it more attractive as an alternative. The superior ratings increase the consumer’s trust level and lower the perceived risk (Ratnasingham 1998), which leads to more purchases of the service leading the market with the highest rating level. Therefore, we hypothesize the following:

Hypothesis 1: A hotel leading the market in ratings will see, on average, a significant increase of its market share.

Endogeneity Issues

Endogeneity is a well-known issue in the econometrics world. A variable is said to be endogenous when there is a correlation between the variable and the error term. It can arise from several reasons: serial autocorrelation, omitted confounding variables, measurement error and bias, and circularity (dynamic model). The model presented in Figure 1 outlines all of the endogeneity issues. Past UGC can influence future UGC and/or market share; variables such as trust are missing by design from a panel data set relying on time series; and property selection can introduce a bias. Our modeling strategy will therefore control for the endogeneity issues by using a dynamic generalized method of moments (GMM) model (Arellano and Bover 1995; Greene 2008) and by testing its pertinence. Therefore, we hypothesize:

Hypothesis 2: There will be a dynamic relationship between market share and UGC.

Ratings and Reviews

Ratings and reviews are expressions of past customers’ experiences. However, we need to make a distinction between ratings and reviews. In line with other research (Dellarocas, Zhang, and Awad 2007), this article considers reviews as a proxy for eWOM. Reviews play a role similar to what WOM plays in the more traditional setting and are sources of information in the consumer decision-making process (Brown, Broderick, and Lee 2007). The impact of online reviews on consumers’ online choices was the topic of research conducted by Sénécal and Nantel (2004). Their results showed that products were selected twice as often when they were recommended. The type of website on which the recommendation is published (i.e., retailer websites versus independent third-party websites) did not have an influence on the perceived trustworthiness and the consumers’ tendency to follow the recommendations. Others have shown the positive impact of online reviews on awareness and purchase consideration (Vermeulen and Seegers 2009).

Ratings for their part can be closely assimilated to an overall service evaluation (Jeong and Jeon 2008). In that sense, the customer uses a single scale to express his or her judgment of the experience. Hence, ratings can be assimilated and used as a proxy of quality as perceived by the past customers. Thus, the ratings viewed by the customer browsing the Internet are a signal of the overall quality to be expected by the reader and should be linked to behavior, such as hotel reservation via the trust customers endow in these ratings. Unfortunately, trust cannot only be measured from each customer browsing the OTA website; hence, it is an omitted variable in our model. Therefore, we hypothesize that ratings are endogenous.

Hypothesis 3: Ratings will be an endogenous variable in the relationship between UGC and market share.

Researchers have found that the relationship between WOM and satisfaction in the context of destination tourism and identity salience is positive (Simpson and Siguaw 2008). Similarly, a relationship between reviews and ratings has been shown (Ye, Law, and Gu 2009). Furthermore, Duan, Gu, and Whinston (2008a) found in a study of online reviews and ratings’ impact on movie revenues that the volume of reviews influences significantly the movie’s revenues. Thus, as can be seen in Figure 1, we suggest that rating, as an expression of satisfaction, not only influences hotel bookings directly but also leads to increased reviews as a form of WOM, which then leads to increased market share by repeat bookings and switching from the competing hotels’ customers. It is hypothesized that reviews will be a confounding variable in the relationship between ratings and market share. Thus,

Hypothesis 4: Reviews will be an endogenous variable in the relationship between UGC and market share.

Lagging Effects

Dellarocas, Zhang, and Awad (2007) found that user rating is the most significant explanatory variable in a revenue forecasting model. Godes and Mayzlin (2004), for their part, focused on measuring the influence of “dispersion” of eWOM, a concept closely related to the awareness effect. Others have shown that awareness is a resulting effect from both positive and negative reviews and explain that even negative reviews might participate in the increase of sales (Xie et al. 2011). Godes and Mayzlin (2004) further examined eWOM communication within and across different user networks and found that dispersion of eWOM is significantly correlated with performance earlier on, while volume exhibits significance only in later periods. These results would suggest that UGC might have a lagging effect on contemporary performance variables such as sales and market share. Thus, we hypothesize:

Hypothesis 5: There will be both contemporary and lagging positive effects of reviews and ratings on market share.

Curvilinear Effect of Rating

Recent findings (Vlachos, Pramatari, and Vrechopoulos 2011) show that service evaluations display a curvilinear relationship with trust. Furthermore, Vlachos, Pramatari, and Vrechopoulos (2011) found in trust a mediator between satisfaction and WOM. The nonlinear effect of satisfaction suggests a zone of tolerance effect (Johnston 1995; Parasuraman, Zeithaml, and Berry 1994; Teas and DeCarlo 2004), by which consumers establish a latitude whereby they accept a variation around an expected level of service. According to the social judgment theory, consumers would assimilate ratings as an expression of satisfaction and quality, falling in the zone of tolerance, to be good and worthy of trust, resulting in booking the hotel; whereby, lower ratings or higher ratings would be rejected. This curvilinear relationship is in line with other service research. Rating, as a reflection of past consumer judgments of quality, does follow a quadratic shape in several findings. For instance, Anderson and Mittal (2000) argue in favor of a nonlinear relationship between satisfaction and retention, which, if not accounted for, would “seriously overestimate the impact of satisfaction” (p. 115). Thus, we hypothesize:

Hypothesis 6: Ratings will exhibit a curvilinear relationship with market share.

Long-Term Effect of UGC

Using elasticity as a measure of short-term and long-term WOM effect, Trusov, Bucklin, and Pauwels (2009) showed that WOM effect, after controlling for endogeneity and autoregressive effects, carry over for several weeks, when traditional marketing activities taper off after a few days. They conclude that researchers should employ models that can also account for the longer-term effect of WOM marketing. Other studies found that ratings might not have, in the long term, a strong effect compared to the short term (Duan, Gu, and Whinston 2008a). Thus, even if the long-term effect seems to be significant, it might be weaker than the short-term effect because of the decay of information over time. Therefore, we hypothesize:

Hypothesis 7: The long-term effect of ratings on market share will be weaker than the short-term effect.

Interaction between Review Length and Valence

Online reviews, when largely ambivalent (positives and negatives), are perceived to be less credible than traditional WOM (Xie et al. 2011). Chevalier and Mayzlin’s (2006) findings suggest there is a strong impact of negative eWOM compared to positive eWOM, because an incremental negative review was more powerful in decreasing book sales than an incremental positive review was in increasing sales.

A distinction between negative and positive eWOM in the eWOM–sales relationship is offered by Samson (2006). His study proposes that negative eWOM affects service firms more than positive eWOM does. An experiment by Laczniak, DeCarlo, and Ramaswami (2001) focused on the influence of negative eWOM on purchase decisions for personal computers. Their results indicated that consumers considered the source of information, particularly negative information, before having a change of opinion about a product or service. Furthermore, recent findings of Duan, Gu, and Whinston (2008b) show a dynamic relationship between eWOM valence, eWOM volume and revenues. This nonlinear relationship could be approached from the framework of the social judgment theory (Sherif and Howland 1961).

According to the theory, consumers would judge reviews’ valence and reviews’ length falling within a zone of tolerance (not too negative [positive] or/and not too lengthy [short]) to be worthy of trust, resulting in an increased chance to book the hotel. The lack of trust associated with the very negative (positive) or very short (long) reviews might also be related to the weak relationship that exists between the third-party consumer and the consumer posting the ratings and reviews (Bansal and Voyer 2000). This relationship can be further supported by the attribution theory and is in line with the findings of Vlachos, Pramatari, and Vrechopoulos (2011) and Chevalier and Mayzlin (2006).

Thus, we suggest that reviews’ valence might interact with reviews’ length, whereby too many words or too many positive reviews might have an opposite effect on trust and diminish the overall impact of UGC on market share. Thus, we hypothesize:

Hypothesis 8: There will be an interaction between review length and review valence that will result in an overall negative effect on market share.

Method

To empirically test the hypotheses, we assembled a database using several sources. First, proprietary data from a sample of hotel properties located in different markets and positioned in different segments were obtained. These data contain monthly market-level performance data over a period of 18 months, as well as a description of the properties’ characteristics. Second, publicly available UGC was extracted from an OTA for each of the properties participating in the study. The sampling strategy and measures are detailed below.

Sample

The sample of hotels for this study is shaped by the following constraints. First, firms must have substantial UGC volume on a monthly basis in order to capture the variability in UGC characteristics needed for the parameters estimation. Second, the segmentation (e.g., star rating) of these firms should be inclusive of all categories (i.e., from economy to luxury), and their brand must be regarded as an inspiration for UGC (Berthon, Pitt, and Campbell 2008). Finally, the study needs longitudinal data on financial metrics used as the dependent variable (i.e., revenue per month for each hotel) and control variables (i.e., market performance, and market concentration) used in the analysis. As such, the present study uses firms located in North America representing branded and independent hotels operating in all market segments (i.e., Luxury, Upper–Midscale, and Economy–Budget) and in all location categories (i.e., Resort, Downtown, Suburbs, and Airport).

Measures and Model Specification

Industry practice uses the RevPAR or Revenue per Available Room index (Rindex) as the metric of choice to measure fair market share (i.e., market share with respect to direct competition). The Rindex is the ratio of the property RevPar to the aggregate competition’s (i.e., the market’s) RevPar. A Rindex of 100 signals a property at par with the market, where a RevPar above (below) 100 suggests a better (worst) performance than the market. Rindex is calculated using the properties’ RevPAR. The RevPAR is the benchmark of the hotel industry performance; it is used by operators, owners, developers, and lodging analysts (Ismail, Dalbor, and Mills 2002). RevPAR captures the interaction of average daily room rate (ADR) and hotel occupancy (OCC) at different phases of the lodging cycle and simultaneously reveals both the supply-and-demand dynamics of a lodging-market cycle in one index (Ismail, Dalbor, and Mills 2002; Woods 1994). The market RevPAR is an important performance measure to compare a property’s RevPAR against. In each market (defined as the property competitive set by management and located within a geographical area around the property) a hotel property that performs well against the market will achieve a RevPAR greater than the market’s RevPAR (the mean RevPAR of all competing hotels). Consequently, a hotel property that underperforms will see its RevPAR fall below the market’s RevPAR. Conceivably, if the market is doing well (i.e., the demand is high with respect to supply) a property that is not marketed well, or that is not managed well, will see its RevPAR moving in the lower range of the market, or at a different pace, possibly with a lag.

Thus, equation (1) models the outcome variable (Rindex_it) RevPar index of property i at time t as a function of the lag RevPar index (Rindex_it-1), UGC_it, a matrix of UGC components, and other control variables (Controls_it) explained below.

\begin{matrix} R i n d e x_{i t} = β_{1} • R i n d e x_{i t - 1} + Β • U G C_{i t} + Γ • C o n t r o l s_{i t} + μ_{i t} \\ where μ_{i t} = ν_{i} + ε_{i t} \end{matrix}

UGC Variables

In equation (1), the UGC matrix is composed of AvgM_it, the mean overall rating for the property i for month t, its lag value AvgM_it-1; AvgM2_it, the square of the mean overall rating for the property i for month t capturing the curvilinear effect; and STAR_itxAvgM_it, which captures the interaction between the segmentation (STAR) of the property and the mean overall rating. Furthermore, Word_it represents the log of the number of words on average in month t; Posit_it, the percentage of positive reviews in month t; and MMkM _it, the relative property rating (AvgM_it) compared to the competition’s mean rating (0: competition does better than property; 1: property does better than competition for month t). Finally, Word_itxPosit_it captures the interaction between the length of the reviews and the average sentiment (percentage of positive vs. negative) of the reviews.

Control Variables

The control variables are LOCA_i, the property i’s location (i.e., Resort, Downtown, Suburbs, or Airport) dummy variable; STAR_i, the segmentation for property i (i.e., Economy–Budget; Upper–Midscale, or Luxury); HHI_i, a measure of the market concentration for property i (i.e., an adapted Herfindahl-Hirschman index for the hotel industry²); DGR_i, the demand growth for property i; DI_i, the demand instability for property i (calculated as the standard deviation of the 5-year average RevPAR growth in the market); and IM_i, which is addressing the potential sample bias using Lee’s (1983) generalization of the Heckman selection correction to create the selection correction variable IM_i (Krasnikov, Jayachandran, and Kumar 2009; Kalaignaman, Shankar, and Varadarajan 2007).³

The Dynamic Model

Several econometric problems may arise from estimating equation (1). First, as shown in our theoretical model (Figure 1), both Rindex_it and UGC_it might be correlated with the error term; therefore, lags need to be modeled in the equation. Second, the variables in UGC_it are assumed to be endogenous and causality runs in both directions, from Rindex_it to UGC_it, and from UGC_it to Rindex_it. Thus, these regressors may also be correlated with the error term. Third, the time-invariant property-specific characteristics (the fixed effects) such as LOCA_i and STAR_i may be correlated with the explanatory variables. The fixed effects are contained in the error term in equation (1), where ν_i is the unobserved effects, and ϵ_it is the observation-specific error. Last, the panel data set has a short time dimension (T = 18) and a larger property dimension (N = 138).

The Arellano–Bond (1991) and Arellano–Bover (1995)/Bundell–Bond (1998) linear generalized method of moments (GMM) dynamic panel estimators are specifically designed for addressing such situations (Bascle 2008; Roodman 2006). Thus, our new model will take the following shape:

\begin{matrix} Δ R i n d e x_{i t} = β_{1} • Δ R i n d e x_{i t - 1} + Β • Δ U G C_{i t} + Γ • Δ C o n t r o l s_{i t} + Δ μ_{i t} \\ where Δ μ_{i t} = Δ ν_{i} + Δ ε_{i t} \end{matrix}

Equation (2) transforms the regressors using first difference. Hence, the time-invariant property-specific effects are removed. To be able to use both lagged Rindex_it and lagged UGC_it as instruments in the level equations, we use a system GMM adding the requirement that E(ΔUGC_itv_it) = 0. In addition, the possibility of UGC_it to be weakly exogenous implies the condition that E(ΔUGC_it-sΔv_it) = 0 for t = 3, …, T and s ≥ 2 (Arellano and Bover 1995).

Results

Description of the Sample and Data

Because of the sensitivity of the data needed (i.e., individual hotel sales over time) across locations, segments, and within markets, the sample strategy was largely driven by the willingness of several anonymous property owners, property management companies, and brand managers to share their Smith Travel Accommodation Report⁴ under confidentiality. One hundred thirty-eight hotels participated, representing all segments and located in 13 markets, of which 22 properties were considered at random to be the group of reference while the rest constituted the competition sets. A postcheck was done to compare the RevPar of the sample (n = 138) to that of the U.S. industry’s overall aggregate for the same period (September 2008 to February 2010). It yielded no apparent eye-balling differences⁵ (see Table 1).

Table 1.

RevPar of Properties Participating and U.S. Average.

Segments^a	Market RevPar of sample	U.S. RevPar (source: STR)
No. of Properties	138	30,022
Economy–Budget	$29.41	$37.13
Upper–Mid Scale	$81.04	$91.82
Luxury	$170.28	$154.11

Note: STR = Smith Travel Research.

Following the hotel industry segmentation with subsegments aggregated. Economy–Budget includes economy and midscale properties; Upper–Mid includes upper-midscale and upper-scale properties; Luxury includes upper-upscale and luxury properties.

Collecting UGC Data

Since the research aim is to disentangle the effect of the different components of UGC on hotel market share, and since theory suggests that consumers’ trust is one of the main antecedents in the UGC–behavior relationship, we sought to get UGC that would be representative of all web-based consumers. As a prior analysis, a sample of 10 properties’ publically available UGC data was extracted directly from the three main OTAs: TripAdvisor.com, Orbitz.com, and Expedia.com, in order to evaluate the convergence of each of these website’s ratings. Using an analysis of variance (ANOVA) method, means of satisfaction (i.e., the mean rating out of 5) for each hotel were compared and a post hoc Sheffé test revealed that the mean difference of Orbitz.com was not significantly different from that of TripAdvisor.com (mean difference = 0.251, SD = 0.17, p = .372) or Expedia.com (mean difference = −0.262, SD = 0.17, p = .342). Thus, we focused solely on Orbitz.com. We concluded that Orbitz.com UGC data are representative of the different web sites allowing consumers to review and rate hotels, and that its policies allow the variance of these ratings to be directly related to actual customer experiences; hence, allowing greater external validity than those of TripAdvisor.com, which accepts noncustomer’s comments.

Descriptive Statistics

Table 2 organizes the data by segments (STAR variable).

Table 2.

Descriptive Statistics by Segment.

Segments	Rindex	SIZE	HHI	DGR	DI	AvgM	AvgMkM	Words	#Mesg	Positive (%)	MMkM (%)
Economy–Budget	130.19	104	1.33	−18.99	12.17	3.47	3.60	15.79	1.53	22.44	37.04
Economy–Budget	(30.28)	(12.74)	(0.47)	(8.61)	(4.57)	(0.68)	(0.31)	(12.25)	(0.63)	22.44	37.04
Upper–Midscale	105.33	238	2	−11.77	6.77	4.25	4.20	40.84	0.63	95.2	38.3
Upper–Midscale	(21.69)	(155.79)	(1.60)	(9.66)	(3.87)	(0.62)	(0.41)	(69.13)	(0.32)	95.2	38.3
Luxury	110.74	370	3.87	−18.40	13.26	4.29	4.38	33.15	1.07	95.3	46.3
Luxury	(21.35)	(215.61)	(3.03)	(6.76)	(6.09)	(0.43)	(0.30)	(35.45)	(0.75)	95.3	46.3

Note: Economy–Budget includes economy and midscale properties; Upper–Mid includes upper-midscale and upper-scale properties; Luxury includes upper-upscale and luxury properties. Rindex = RevPar index; SIZE = number of rooms; HHI = concentration index; DGR = demand growth; DI = demand volatility; AvgM = average rating; AvgMkM = average competition rating; Words = number of words per review; #Mesg = number of reviews per month; Positive = proportion of positive reviews; MMkM = proportion of time above market rating.

We notice that the RevPar index of the sample is 30% greater than that of the competitors in the data set for the economy-budget segment (Rindex = 130), followed by the Luxury segment with an 11% Rindex above competition, and finally the Upper–Midscale segment showing a 5% increase over competition. Although it appears that the sample is biased toward successful properties, it should be noted that these results are aggregated over each segment and that their standard deviations imply that the confidence interval would include 100 RevPar index or parity with competition in close to 50% of the sample. Of notable importance, in Table 2, we notice that the market concentration (HHI) of Economy–Budget hotels tends to lean toward a more competitive setting (HHI = 1.33, SD = 0.47) than for the Luxury market (HHI = 3.87, SD = 3.03). Demand growth (DGR) seems to reflect the recessionary pattern of the period under study (September 2008 to February 2010), although the Upper–Midscale segment seems to fare better (DGR of −11.77, SD = 9.66, compared to a DGR of −18.99, SD = 8.61, for the Economy–Budget segment and −18.40, SD = 6.76, for the Luxury segment).

Additionally, the demand instability (DI) or volatility in growth confirms that the Upper–Midscale segment is faring somewhat better than the other two with almost half a DI value on average.

UGC Results by Segment

The first element of UGC is the mean rating of the properties (AvgM) compared to the competitors’ mean ratings (AvgMkM). Although Economy–Budget and Luxury segments have slightly lower AvgM than their competition, and the Upper–Midscale segment has a slightly higher AvgM than competition, the differences once tested statistically were not significant (t _{Eco-Budget, df=136} = 1.43; t _{Upper-Mid, df=136} = 0.48; t _{Luxury, df=136} = 1.20). However, differences between segments were found to be significant between the Eco-Budget segment and the Upper–Midscale and Luxury segments (t _df=117 = 7.33). No significant difference was uncovered between the Upper–Midscale and the Luxury segments (t _df=106 = 1.81). This would imply that segmentation does play a role in the rating of properties and that a gap in ratings exists between the lowest level (Eco-Budget) and the highest (Upper–Midscale and Luxury). These results are also in accordance with results obtained by research firms across segments (see J.D. Power and Associates Reports 2011 and ACSI 2011).

The number of words on average per message per segment suggests that the Economy–Budget customers tend to write fewer words than the Upper–Midscale or Luxury customers. However, it seems that the most prolific customers reside in the Upper–Midscale segment, where 41 words on average per message is used with a large variability (SD = 69.13) within the group. The valence of the messages, or whether they are positive or negative, shows that the Upper–Midscale and Luxury segment customers use more positive messages (95%) than the Economy–Budget customers (22%) on average.

Estimation of the Model

Using equation (2), a system GMM model is estimated taking into account possible heterogeneity, endogeneity, autoregression, and the dynamic aspect of the panel data (see Table 3).

Table 3.

System GMM Model of UGC on Hotel Market Share.

Dependent Variable: Rindex	Coefficient	t Value	p Value
Intercept	−9.999	−0.49	ns
Rindex_t _–1	0.590	25.53	<.001
UGC effects
AvgM	86.482	6.53	<.001
AvgM_t _–1	6.085	1.92	<.10
AvgM2	−12.785	−7.82	<.001
AvgMStar*	17.970	7.81	<.001
Word	−27.962	−7.92	<.001
Posit	−102.529	−5.76	<.001
MMkM	−0.412	−0.34	ns
PW	26.614	6.57	<.001
Controls
LOCA	−3.894	−2.41	<.05
Star	−87.743	−8.6	<.001
HHI	−0.003	−0.48	ns
DGR	−0.432	−1.3	ns
DI	0.393	0.66	ns
IM	11.179	7.87	<.001
Fit indices
R²/pseudo-R²	74.5%
F test	249.38		<.001
No. of observations	1374
No. of groups	171
No. of instruments	139
AR (1)	z = −2.86		<.001
AR (2)	z = −1.33		ns
Hansen test	χ²(123) = 149.33
	p = 0.053
Difference-in-Hansen test p-value	0.58

Note: ns = non-significant; GMM-style instruments are Rindex and AvgM using lag 3 as starting point. All variables are in first differences.

Our most likely appropriate GMM model treats Rindex_it and the UGC_it variables as endogenous using lag 3 as a starting point. All the control variables are considered strictly exogenous. The AR(1) term is significant, AR(1) = −2.86, p < .001, as predicted, but yields a nonsignificant AR(2) effect, AR(2) = −1.33, p > .05. The results from the GMM pass the instrument validity test (Hansen test = 149.33, p = .053; difference in Hansen test, p = .58) suggesting an appropriate model (Greene 2008).

The GMM model explains 74.5% of the total variance in Rindex_it. As expected, a large portion of the variance is explained by the lag variable (Rindex_it _-1), showing that 59% of the average hotel RevPar index is simply due to the past month’s business. The control variables’ estimates are significant, with the exception of HHI_i, DGR_i, and DI_i. This could indicate that market conditions such as concentration, growth, and volatility do not significantly impact hotel market shares relative to other competing hotels. In other words, all hotels located in the same market suffer the same way from market conditions. The selection correction variable (IM_i) is positive and significant (IM_i = 11.18, p < .001), underscoring the need to control for selection bias. We now turn to the UGC_it estimates and the hypotheses testing.

Results of Hypotheses Testing

Evidenced by the good-fit GMM model, we can be confident that the variables exhibit a dynamic relationship with endogeneity in the dependent variable, as well as the UGC variables, and a lag effect; thus, showing support to hypothesis 2 (There will be a dynamic relationship between market share and ratings and reviews), as well as hypotheses 3 and 4 (Ratings and reviews will be an endogenous variable in the relationship between UGC and market share).

A hotel that would get ratings, on average, above its competitive set (MMkM_it), would not fare better than a hotel that would be consistently below its competition. In fact, the estimate for MMkM_it is not different from zero (MMkM_it = −0.412, t = −0.34). Thus, hypothesis 1 (A hotel leading the market in ratings will see, on average, a significant increase of its market share) is not supported.

The positive parameter estimate for the effect of AvgM_it on Rindex_it in the final model (AvgM _it = 86.48, t = 6.53) demonstrates that hotels with higher AvgM_it will tend to secure higher market shares, thus supporting hypothesis 5 (There will be both contemporary and lagging positive effects of reviews and ratings on market share). Furthermore, AvgM_it-1 estimate is positive and significant (AvgM_it _-1 = 6.08, t = 1.92) suggesting that past rating affects current market share. This is not surprising since the first page of OTA shows a cumulative running rating for each hotel, where consumers have to click on the particular hotel in order to view individual comments and ratings.

The estimate for AvgM2_it is negative and significant (AvgM2_it = −12.79, t = −7.82), indicating a quadratic relationship between AvgM_it and Rindex_it, supporting hypothesis 2 (The relationship between the service firm’s ratings and its market share will show diminishing return). Furthermore, the estimate for the cross-level interaction effect of segmentation on rating (STAR_ixAvgM_it) is positive and significant (STAR_ixAvgM_it = 18.90, t = 2.34), indicating that the maximum for the effect of AvgM_it is increasing as hotels are positioned in higher tiers, showing support for hypothesis 6 (Ratings will exhibit a curvilinear relationship with market share). Table 4 displays the maxima by segment (STAR). Hotels competing in the Economy–Budget segment might be able to trade quality for profit since a rating of only 3.62 out of 5, or 72%, would not affect market share negatively, but a rating higher than 3.62 would. Hotels in the Midscale segment will see market share decline around a maximum rating of 4.32 out of 5, or 87%. As would be expected, hotels competing in the Luxury segment must strive for the highest ratings. It seems that short of a perfect 5 out of 5 Luxury properties would be doomed to lose market share to the competition.

Table 4.

Rating Maximum per Segment.

Segments	Economy–Budget	Upper–Mid	Luxury
Mean rating out of 5	3.62	4.32	5.03
As a percentage	72.4%	86.5%	100.5%

Using the Papke and Wooldridge (2004) formula, we obtained both long-run estimates and standard errors for the dynamic effects as shown in Table 5. The long-term effect of AvgM_it on Rindex_it is significantly greater than the short-term effect for all segments. Thus, hypothesis 7 (The long-term effect of ratings on market share will be weaker than the short-term effect) is not supported.

Table 5.

Long Run Effect of Ratings on Market Share.

	Economy–Budget	Upper–Mid	Luxury
Short run	3.620	4.323	5.025
	(0.086)	(0.098)	(0.181)
Long run	8.839	10.555	12.271
	(0.468)	(0.597)	(0.803)

Note: Estimates and (standard errors) per segment.

As positive reviews increase as a percentage of all reviews, market share tends to decrease (Posit_it = −102.53, t = −5.76). Similarly, reviews that tend to have more words tend to decrease market share on average (Word_it = −27.96, t = −7.92). The interaction term between review valence and length (Posit_it × Word_it) is positive and significant (Posit_it × Word_it = 26.61, t = −6.57). Using a low (high) value of 5 (100) words for Word, and a low (high) value of 25% (75%) for Posit_it, we can estimate that the net impact on market share, including the interaction effect when Word_it is high but Posit_it is low or high, seems fairly equivalent (low Posit–high Word = −113 vs. high Posit–high Word = −124). However, when Word_it is low, negative reviews will amplify the negative impact on market share (high Posit–low Word = −30 vs. low Posit–low Word = −60) in the magnitude of 100%. Thus, overall reviews’ valence and length impact market share negatively. Positive reviews associated with fewer words will reduce the negative impact on market share, but not enough to be positive. The more words in a review whether positive or negative, the more negative effect it has on market share. The overall negative effect of Word and Posit on market share supports hypothesis 8 (There will be an interaction between review length and review valence that will result in an overall negative effect on market share).

Discussion and Implications

Overall, our analysis reveals that UGC, as a whole, has a positive impact on market share. However, several counterintuitive results make UGC an important marketing metrics to monitor and to integrate as part of an overall customer relationship management initiative. Ratings have been shown to exhibit a curvilinear relationship with market share, more pronounced in the lower-tiered segments than the upper-tier or luxury ones. More surprising are the findings that reviews no matter how good they can be for the property, seem to have a negative impact overall on market share less when they are positive than when they are negative. Furthermore, properties that would lead the competition with respect to their ratings would not necessarily see a benefit in market share. All results seem to point to the fact that UGC, per segment, has a “sweet spot” where maximum effect on market share is obtained, but causes the reverse effect when they are too good to be true. The issue at the core of the relationship between UGC and market share seems to be consumers’ trust level and choice mechanism when using OTAs.

Theoretical Contributions

Our study makes several contributions to the marketing theory. First, it addresses researchers’ recent calls to use qualitative research methods to inform longitudinal analysis (Banyai and Glover 2012) and to link UGC, not only to customer satisfaction and delight (Crotts, Mason, and Davis 2009; Magnini, Crotts, and Zehrer 2011) but also to consumer behavior (Litvin, Goldsmith, and Pan 2008) using nonlinear relationships (Rust, Danaher, and Varki 2000). Second, several recent pieces of research have linked UGC to sales of products or movies, yet no study has linked UGC to firms’ market share in a longitudinal panel-data design. Third, and unlike prior research that provided only partial insights into the effects of UGC on product-level sales (Ye, Law, and Gu 2009), the current study extends the finding to show the impact of UGC relative to competition and not simply as a correlation to sales. Finally, in the empirical context of hotels, the study shows that UGC characteristics act differently on market share than other studies have found. Specifically, review volume and valence impact market share to a lesser degree than ratings. This finding is contrary to other research, but it is important to note that these researches were conducted in different empirical settings (Dellarocas, Awad, and Zhang 2006; Duan, Gu, and Whinston 2008a; Liu 2006). The empirical setting might be the underlying reason for the differences in results. Following are different possible explanations for our results along with theoretical support.

Ratings for Economy and Budget hotels culminate at 3.62 out of 5 (72.4%) with a quadratic relationship to market share. There is evidence of a shift toward higher maximum as the segmentation level increases. The curvilinear relationship shown between ratings and market share is known in service research. Rating as a reflection of past consumer judgments of quality has been shown to follow a quadratic shape (Anderson and Mittal 2000; Voss et al. 2004; Zeithaml, Parasuraman, and Malhotra 2002). The curvilinear effect of satisfaction involves a zone of tolerance (Parasuraman, Zeithaml, and Berry 1994) and can also be explained in light of the social judgment theory (Sherif and Howland 1961), particularly in the context of judgment made by the third party (i.e., the consumer reading the ratings of another consumer and evaluating its validity). Consistent with the social judgment theory, the latitude of rejection is greater toward the low end of the scale (between a rating of 1 and 3.62 for Economy–Budget hotels), than it is toward the high end of the scale (between a rating of 3.62 and 5). The perceived risk of negative reviews or low ratings might explain the negative skewness of the quadratic curve (Murray 1991). The relationship between perceived risk and the rating maximum is evidenced in the Luxury segment since the peak of the curve equates to the maximum rating, leaving no room for error and justifying the heavy emphasis on quality.

The lack of trust associated with the high ratings beyond the maximum in the Economy–Budget and Upper–Midscale segments might be related to the weak relationship that exists between the third-party consumer and the consumer posting the ratings and reviews (Bansal and Voyer 2000). In a many-to-many communication environment (Hoffman and Novak 1996), interpersonal relationships between sender and receiver are not strong ones (Brown, Broderick, and Lee 2007; Parks and Floyd 1996) and the expertise level of the customer posting reviews and ratings might be questioned (Feick and Higie 1992). Although consumers have been known to demonstrate a level of trust in using product reviews (Riegner 2007), reviews and ratings of service experiences might suffer from the subjective nature of the assessment and the judgment made by the reader (Laczniak, DeCarlo, and Ramaswami 2001; Folkes 1988; Sen and Lerman 2007; Weiner 2000).

In light of the interactions in the model, and along with current research (Laczniak, DeCarlo, and Ramaswami 2001; Samson 2006), our study found that the length of the reviews is important especially when the review is negative. Our results show that although a long negative review has a similar impact as a long positive one does, short reviews affect market share less negatively, and positive short reviews have the least negative impact. We explain these results in light of the attribution theory (Folkes 1988; Weiner 2000). Consumers, especially in the service context, attribute the extreme reviews to the reviewers’ personal level of expectation, and not necessarily to any locus of control or fault of the service provider.

Alternative explanations for our results could be explained in light of the low level of involvement a customer would exhibit when purchasing a hotel room (Johnston 1995). It is proposed that within a low level of involvement, satisfaction rather than delight, might be enough; positive transactions might not be required and negative ones can be compensated (Johnston 1995). This explanation is also in line with noncompensatory decision rules tendencies in the service industry (Rust, Danaher, and Varki 2000), which would suggest that heuristic rules are used when consumers are limited in their processing time. For instance, our results could suggest that the lexicographic rule be the underlying heuristic when choosing a hotel online. The most important attribute (the rating relative to the zone of tolerance) would prevail in the choice, except if the alternatives are tied; in which case the next attribute (reviews) would be evaluated, although one more “click” is needed for consumers to read reviews, hence requiring a deeper level of involvement. Further research could investigate models that would account for click-through behaviors in order to explain purchases.

Managerial Implications

This study has several useful implications for managers. First, managers should concentrate on quality to drive ratings. However, given the managerial pressure, service failures are bound to happen, triggering lower ratings and negative reviews; thus, the issue at large is more about complaint resolution and postservice management relation than it is about control, at least in the lower-tiered segments. Managers of Economy–Budget and Upper–Midscale hotels would want to encourage ratings and reviews, and would want to mitigate, rather than eliminate, the negative reviews. Unless operating in the Luxury segment, overly positive and lengthy reviews might be seen as suspect by consumers; hence, managers should not fear to be rated “in the middle” of the tolerance zone.

Consistent with some popular press opinion (Yu 2010), our results empirically demonstrate that on average “70% quality is good enough.” This is true for the lowest-tiered segments, but the Upper–Midscale segment should strive for more (87.6%) and the Luxury for perfection. These levels of quality are consistent with the national average by segments. Unless located in the Luxury segment, managers might deal with internal and external pressures requiring different goals. For instance, corporate headquarters might request 100% quality as a nonnegotiable goal in order to be able to use the rating as a public relations talking point. Corporate headquarters might even go farther and set incentives for individual properties that are linked to the achievement of high levels of customer ratings. This is counterbalanced by the tendency of property owners to go in the opposite direction, as they often see quality as a cost and not as an investment. In light of our results, and independently of the public relations problem, goals and incentives should foster a system similar to that used in sales. For example, an Economy–Budget hotel manager that achieves a 72% rating might currently be sanctioned. However, that hotel rating is in the zone of tolerance and should be rewarded. Furthermore, reviews should be quantified for length and valence, possibly by a third-party vendor using an algorithm. Results could be incorporated in the monthly benchmarking analysis firms.

Even if an OTA represents only about 7% of the hotel’s business booked online (Hood 2011), management should take OTAs seriously and consider the “billboard effect,” where consumers have been shown to visit between seven and eight OTAs before booking their hotel, and even then might end up using the hotel chain’s website to make their purchase in the end (Green 2011). Meanwhile, a study reports that two of every three online traveler reviews are posted to an OTA website in 2010 (Green 2011). Therefore, favorable UGC not only produces bookings via the OTAs but also can produce bookings directly, generating higher margins.

Limitations and Future Research Directions

This study may suffer from several limitations. First, even if our sample of hotels show robustness when compared to the national average, it is possible that even while controlling for selection bias, our results will not be generalizable. Second, the empirical setting of hotels might be unique and not representative of purely web-based services or click-and-mortar retailers. Hence, these limitations can provide guidelines for future research.

Studies that will expand the empirical setting to other services presenting similar characteristics such as restaurants would help the generalizability of the results. Similarly, studies that will examine services in the context of high involvement (i.e., investments, car or house purchases, hospitals) would contribute to our understanding of the UGC phenomenon. As mentioned earlier, studies that would further our understanding of the relationship between click-through behavior and purchase decision, in the context of reviews and ratings, would also be of great interest to the field of marketing in general. Finally, studies that would integrate offline promotions in the model would further the understanding of the dynamic relationship off- and online.

Footnotes

Acknowledgements

The author expresses appreciation to the anonymous reviewers, the Editor, Dr. Garland Keesling, Steve Hood at Smith Travel Research, and all the hoteliers who aided in the development of this research project.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biography

Philippe Duverger has a PhD from George Washington University and is an assistant professor of marketing in the College of Business and Economics at Towson University. His research focuses on service innovation and quantitative methods. Prior to obtaining his PhD, he spent 20 years in the hospitality and service industry in different executive capacities and is a Certified Hotel Administrator (CHA).

References

ACSI. (2011). Historical ACSI Scores. http://www.theacsi.org/index.php?option=com_content&view=article&id=213:acsi-scores-june&catid=14&Itemid=262 (accessed June 21, 2011).

Anderson

E. W.

Mittal

(2000). “Strengthening the Satisfaction-Profit Chain.” Journal of Service Research, 3 (2): 107-20.

Arellano

Bond

S. R.

(1991). “Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations.” Review of Econometric Studies, 58: 277-97.

Arellano

Bover

(1995). “Another Look at the Instrumental Variable Estimation of Error Component Models.” Journal of Econometrics, 68: 29-51.

Bansal

H. S.

Taylor

S. F.

St. James

(2005). “‘Migrating’ to New Service Providers: Toward a Unifying Framework of Consumers’ Switching Behaviors.” Journal of the Academy of Marketing Science, 33 (1): 96-115.

Bansal

H. S.

Voyer

P. A.

(2000). “Word-of-Mouth Processes within a Services Purchase Decision Context.” Journal of Service Research, 3 (2): 166-77.

Banyai

Glover

T. D.

(2012). “Evaluating Research Methods on Travel Blogs.” Journal of Travel Research, 51 (3): 267-77.

Bascle

(2008). “Controlling for Endogeneity with Instrumental Variables in Strategic Management Research.” Strategic Organization, 6: 285-327.

Berthon

Pitt

Campbell

(2008). “Ad Lib: When Customers Create the Ad.” California Management Review, 50 (Summer): 6-30.

10.

Bronner

de Hoog

(2010). “Vacationers and eWOM: Who Posts, and Why, Where, and What?” Journal of Travel Research, 50 (1): 15-26.

11.

Brown

Broderick

A. J.

Lee

(2007). “Word of Mouth Communication within Online Communities: Conceptualizing the Online Social Network.” Journal of Interactive Marketing, 21: 2-20.

12.

Bundell

Bond

S. R.

(1998). “Initial Conditions of Moment Restrictions in Dynamic Panel Data Models.” Journal of Econometrics, 87 (1): 115-43.

13.

Chen

Xie

(2008). “Online Consumer Review: A New Element of Marketing Communications Mix.” Management Science, 54 (3): 477-91.

14.

Chevalier

J. A.

Mayzlin

(2006). “The Effect of Word of Mouth on Sales: Online Book Reviews.” Journal of Marketing Research, 43 (3): 345-54.

15.

Cox

Burgess

Sellitto

Buultjens

(2009). “The Role of User-Generated Content in Tourists’ Travel Planning Behavior.” Journal of Hospitality Marketing and Management, 18 (8): 743-64.

16.

Crotts

J. C.

Mason

P. R.

Davis

(2009). “Measuring Guest Satisfaction and Competitive Position in the Hospitality and Tourism Industry: An Application of Stance-Shift Analysis to Travel Blog Narratives.” Journal of Travel Research, 48 (2): 139-51.

17.

Dellarocas

Awad

F. N.

Zhang

(2006). “Using Online Rating as a Proxy of Word-of-Mouth in Motion Picture Revenue Forecasting.” Working Paper.

18.

Dellarocas

Zhang

Awad

N. F.

(2007). “Exploring the Value of Online Product Reviews in Forecasting Sales: The Case of Motion Pictures.” Journal of Interactive Marketing, 21 (4): 23-45.

19.

De Lollis

(2012). “Best Western Latest to Highlight Hotel Reviews on Website,” USA Today, September 25, http://travel.usatoday.com/hotels/post/2012/09.

20.

De Matos

C. A.

Rossi

C. A. V.

(2008). “Word-of-Mouth Communications in Marketing: A Meta-analytic Review of the Antecedents and Moderators.” Journal of the Academy of Marketing Science, 36: 578-96.

21.

Dickinger

(2011). “The Trustworthiness of Online Channels for Experience- and Goal-Directed Search Tasks.” Journal of Travel Research, 50 (4): 378-91.

22.

Duan

Whinston

A. B.

(2008a). “Do Online Reviews Matter?—An Empirical Investigation of Panel Data.” Decision Support Systems, 45: 1007-16.

23.

Duan

Whinston

A. B.

(2008b). “The Dynamics of Online Word-of-Mouth and Product Sales—An Empirical Investigation of the Movie Industry.” Journal of Retailing, 84 (2): 233-42.

24.

Feick

Higie

R. A.

(1992). “The Effects of Preference Heterogeneity and Source Characteristics on Ad Processing and Judgments about Endorsers.” Journal of Advertising, 21 (2): 9-24.

25.

Folkes

V. S.

(1988). “Recent Attribution Research in Consumer Behavior: A Review and New Directions.” Journal of Consumer Research, 14 (4): 548-65.

26.

Godes

Mayzlin

(2004). “Using Online Conversations to Study Word-of-Mouth Communication.” Marketing Science, 23 (4): 545-60.

27.

Green

C. E.

(2011). Distribution Strategies Issues, Smith Travel Research, Hotel Data Conference, Nashville, August 3-4, 2011.

28.

Greene

W. H.

(2008). Econometric Analysis. 6th edition. Upper Saddle River, NJ: Pearson International.

29.

H. Y.

(2004). “Factors Influencing Consumer Perceptions of Brand Trust Online.” Journal of Product & Brand Management, 13 (5): 329-42.

30.

Harris

L. C.

Goode

M. M. H.

(2004). “The Four Levels of Loyalty and the Pivotal Role of Trust: A Study of Online Service Dynamics.” Journal of Retailing, 80 (2): 139-58.

31.

Hoffman

D. L.

Novak

T. P.

(1996). “Marketing in Hypermedia Computer-Mediated Environments: Conceptual Foundations.” Journal of Marketing, 60 (3): 50-68.

32.

Hood

(2011). Distribution Channels Analysis, Smith Travel Research, Hotel Data Conference, Nashville, August 3-4, 2011.

33.

Interactive Advertising Bureau. (2008). “User Generated Content, Social Media, and Advertising—An Overview.” IAB 2008 UGC and Social Media Event (April). http://www.iab.net/events_training/326160 (accessed December 10, 2008).

34.

Ismail

J. A.

Dalbor

M. C.

Mills

J. E.

(2002). “Lodging-Segment Variability: Using RevPAR to Analyse.” Cornell Hotel and Restaurant Administration Quarterly, 43: 73-80.

35.

J. D. Power and Associates Reports. (2011). Hotel Guest Satisfaction with Costs and Fees Improves from 2010, Despite a Decline in Overall Satisfaction. http://businesscenter.jdpower.com/news/pressrelease.aspx?ID=2011106.

36.

Jeong

Jeon

M. M.

(2008). “Customer Reviews of Hotel Experiences through Consumer Generated Media.” Journal of Hospitality Marketing and Management, 17 (1/2): 121-38.

37.

Johnston

(1995). “The Zone of Tolerance: Exploring the Relationship between Service Transactions and Satisfaction with the Overall Service.” International Journal of Service Industry Management, 6 (2): 46.

38.

Jones

M. A.

Mothersbaugh

D. L.

Beatty

S. E.

(2000). “Switching Barriers and Repurchase Intentions in Services.” Journal of Retailing, 76 (2): 259-74.

39.

Kalaignaman

Shankar

Varadarajan

(2007). “Asymmetric Product Development Alliances: Win-Win or Win-Lose Partnerships?” Management Science, 53 (3): 357-74.

40.

Kirkpatrick

Roth

(2005). “Why There’s No Escaping the Blog.” Fortune, January 10, 44-50.

41.

Krasnikov

Jayachandran

Kumar

(2009). “The Impact of Customer Relationship Management Implementation on Cost and Profit Efficiencies: Evidence from the U.S. Commercial Banking Industry.” Journal of Marketing, 73 (6): 61-76.

42.

Laczniak

R. N.

DeCarlo

T. E.

Ramaswami

S. N.

(2001). “Consumers’ Responses to Negative Word-of-Mouth Communication: An Attribution Theory Perspective.” Journal of Consumer Psychology, 11 (1): 57-73.

43.

Lee

(1983). “Generalized Econometric Models with Selectivity.” Econometrica, 51 (2): 507-12.

44.

Litvin

S. W.

Goldsmith

R. E.

Pan

(2008). “Electronic Word-of-Mouth in Hospitality and Tourism Management.” Tourism Management, 29 (3): 458-68.

45.

Liu

(2006). “Word of Mouth for Movies: Its Dynamics and Impact on Box Office Revenue.” Journal of Marketing, 70 (July): 74-89.

46.

Magnini

V. P.

Crotts

J. C.

Zehrer

(2011). “Understanding Customer Delight: An Application of Travel Blog Analysis.” Journal of Travel Research, 50 (5): 535-45.

47.

McDougall

G. H. G.

Levesque

(2000). “Customer Satisfaction with Services: Putting Perceived Value into the Equation.” Journal of Services Marketing, 14 (5): 392-410.

48.

Murray

K. B.

(1991). “A Test of Services Marketing Theory: Consumer Information Acquisition Activities.” Journal of Marketing, 55 (1): 10-25.

49.

O’Connor

(2010). “Managing a Hotel’s Image on TripAdvisor.” Journal of Hospitality Marketing and Management, 19 (7): 754-72.

50.

Papke

L. E.

Wooldridge

J. M.

(2004). “A Computational Trick for Delta-Method Standard Errors.” Economics Letters, 86: 413-17.

51.

Parasuraman

Zeithaml

V. A.

Berry

L. L.

(1994). “Reassessment of Expectations as a Comparison Standard in Measuring Service Quality: Implications for Further Research.” Journal of Marketing, 58 (1): 111-24.

52.

Parks

M. R.

Floyd

(1996). “Making Friends in Cyberspace.” Journal of Communication, 46: 80-97.

53.

Ratnasingham

(1998). “The Importance of Trust in Electronic Commerce.” Internet Research: Electronic Networking Applications and Policy, 8 (4): 313-21.

54.

Riegner

(2007). “Word of Mouth on the Web: The Impact of Web 2.0 on Consumer Purchase Decisions.” Journal of Advertising Research, 47 (4): 436-47.

55.

Roodman

(2006). “How to Do Xtabond2: An Introduction to ‘Difference’ and ‘System’ GMM in Stata.” Center for Global Development Working Paper Number 103.

56.

Rust

R. T.

Danaher

P. J.

Varki

(2000). “Using Service Quality Data for Competitive Marketing Decisions.” International Journal of Service Industry Management, 11 (5): 438-69.

57.

Samson

(2006). “Understanding the Buzz That Matters: Negative vs Positive Word of Mouth.” International Journal of Market Research, 48 (6): 647-57.

58.

Sen

Lerman

(2007). “Why Are You Telling Me This? An Examination into Negative Consumer Reviews on the Web.” Journal of Interactive Marketing, 21 (4): 76-94.

59.

Sénécal

Nantel

(2004). “The Influence of Online Product Recommendations on Consumers’ Online Choices.” Journal of Retailing, 80: 159-69.

60.

Sharkey

(2010). “Named and Shamed: The World’s Dirtiest Hotels.” The New York Times, February 13.

61.

Sherif

Howland

(1961). Social Judgment: Assimilation and Contrast Effects in Communication and Attitude Change. New Haven, CT: Yale University Press.

62.

Simpson

P. M.

Siguaw

J. A.

(2008). “Destination Word of Mouth.” Journal of Travel Research, 47 (2): 167-82.

63.

Smith

Menon

Sivakumar

(2005). “Online Peer and Editorial Recommendations, Trust, and Choice in Virtual Markets.” Journal of Interactive Marketing, 19 (3): 15-37.

64.

Sparks

B. A.

Browning

(2011). “The Impact of Online Reviews on Hotel Booking Intentions and Perception of Trust.” Tourism Management, 32: 1310-23.

65.

Stringham

B. B.

Gerdes

Jr. Vanleeuwen

D. M.

(2010). “Assessing the Importance and Relationships of Ratings on User-Generated Traveler Reviews.” Journal of Quality Assurance in Hospitality and Tourism, 11 (2): 73-92.

66.

Susskind

A. M.

Bonn

M. A.

Dev

C. S.

(2003). “To Look or Book: An Examination of Consumers’ Apprehensiveness toward Internet Use.” Journal of Travel Research, 41 (3): 256-64.

67.

Teas

R. K.

DeCarlo

T. E.

(2004). “An Examination and Extension of the Zone-of-Tolerance Model.” Journal of Service Research, 6 (3): 272-86.

68.

Trusov

Bucklin

R. E.

Pauwels

(2009). “Effects of Word-of-Mouth versus Traditional Marketing: Findings from an Internet Social Networking Site.” Journal of Marketing, 73 (September): 90-102.

69.

Vermeulen

I. E.

Seegers

(2009). “Tried and Tested: The Impact of Online Hotel Reviews on Consumer Consideration.” Tourism Management, 30: 123-27.

70.

Verna

(2008). “User-Generated-Content: In Pursuit of Ad Dollars.” eMarketer Analyst Report, April, 35.

71.

Vlachos

P. A.

Pramatari

Vrechopoulos

(2011). “Too Much of a Good Thing: Curvilinear Effects of Service Evaluation Constructs and the Mediating Role of Trust.” Journal of Services Marketing 25 (6): 440-50.

72.

Voss

C. A.

Roth

A. V.

Rosenzweig

E. D.

Blackmon

Chase

R. B.

(2004). “A Tale of Two Countries’ Conservatism, Service Quality, and Feedback on Customer Satisfaction.” Journal of Service Research, 6: 212-30.

73.

Weiner

(2000). “Attributional Thoughts about Consumer Behavior.” Journal of Consumer Research, 27 (3): 382-87.

74.

Woods

Jr. (1994). “Using Rooms-Revenue-Per-Available-Room Analysis to Forecast Rooms Revenue.” Real Estate Finance Journal, Spring: 20-29.

75.

Xie

Miao

Kuo

Lee

(2011). “Consumer’ Responses to Ambivalent Online Hotel Reviews: The Role of Perceived Source Credibility and Pre-decisional Disposition.” International Journal of Hospitality Management, 30: 178-83.

76.

Law

(2009). “The Impact of Online User Reviews on Hotel Room Sales.” International Journal of Hospitality Management, 28: 180-82.

77.

(2010). “Hotel Managers Monitor Online Critiques to Improve Service.” USA Today, March 23.

78.

Zeithaml

V. A.

Parasuraman

Malhotra

(2002). “Service Quality Delivery through Web Sites: A Critical Review of Extant Knowledge.” Journal of the Academy of Marketing Science, 30 (4): 362-75.

79.

Zhu

Zhang

(2010). “Impact of Online Consumer Reviews on Sales: The Moderating Role of Product and Consumer Characteristics.” Journal of Marketing, 74 (2): 133-48.